Using Natural Language Processing to Automatically Detect Self-Admitted Technical Debt
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The metaphor of technical debt was introduced to express the trade off between productivity and quality, i.e., when developers take shortcuts or perform quick hacks. More recently, our work has shown that it is possible to detect technical debt using source code comments (i.e., self-admitted technical debt), and that the most common types of self-admitted technical debt are design and requirement debt. However, all approaches thus far heavily depend on the manual classification of source code comments. In this paper, we present an approach to automatically identify design and requirement self-admitted technical debt using Natural Language Processing (NLP). We study 10 open source projects: Ant, ArgoUML, Columba, EMF, Hibernate, JEdit, JFreeChart, JMeter, JRuby and SQuirrel SQL and find that 1) we are able to accurately identify self-admitted technical debt, significantly outperforming the current state-of-the-art based on fixed keywords and phrases; 2) words related to sloppy code or mediocre source code quality are the best indicators of design debt, whereas words related to the need to complete a partially implemented requirement in the future are the best indicators of requirement debt; and 3) we can achieve 90 percent of the best classification performance, using as little as 23 percent of the comments for both design and requirement self-admitted technical debt, and 80 percent of the best performance, using as little as 9 and 5 percent of the comments for design and requirement self-admitted technical debt, respectively. The last finding shows that the proposed approach can achieve a good accuracy even with a relatively small training dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it